The field of open-domain conversation generation using deep neural networks has attracted increasing attention from researchers for several years. However, traditional neural language models tend to generate safe, generic reply with poor logic and no emotion. In this paper, an →